Weighted Combination of Naive Bayes and LVQ Classifier for Fongbe Phoneme Classification

In speech recognition, phoneme classification has recently gained increased attention. The combination of classifiers has emerged as a reliable method and is used for decision-making by combining individual opinions to produce a final decision. In this study, we propose a novel classifier based on the combination of Naive Bayes and Learning Vector Quantization (LVQ) using weighted voting to recognize the consonants and vowels of a local language Fongbe in Benin. Indeed we are faced with a problem of lack of training data where the results of different classifiers may be uncertain. To improve decisions, in this work we combine a classification approach based on probability theory and another approach based on finding the nearest neighbor. Different techniques of speech analysis are used for evaluation and results show that the most significant classification rates were achieved with PLP coefficients. The different results showed the effectiveness of our approach.